[Home]DarkHallow/WhatLessonHere

Robo Home | DarkHallow | Changes | Preferences | AllPages

Showing revision 8
In the last two days I have reworked my surfing to work with a more accurate projection system. In the process I have stayed relatively flat in the ratings. So what I am wondering now is what lesson is there to be taken away from this? Some of my initial thinking:

Perhaps more importantly where do I go from here? I am at a loss?


I think strategery (as our beloved president would say) is probably more important than precise prediction. There are very few bots where you can exactly predict where their bullets will be, and knowing exactly where you will be only helps in those cases. --David Alves

I think precise, bloody precise, prediction is the way to go. Any tactics you decide on trying will depend on this. It's no coincidence that the bots having reached 2035 points all, with one exception, are precise predictors. I know I was able push CC past 2050 without the last little piece of preciseness. But it came with a price on my sanity. I tried to push forward everywhere without any sensible results. Some bugs I removed lowered my rankings and insane experiments sometimes payed off... Once I had precise prediction in place, reasonably bug free, CC started to respond to my tweaks in somewhat expected ways at least. Now, when DH has precise prediction in place, you probably must go further and implement non-fuzzy dive-in-protection (evaluation of the stop position together with refusal to smooth too hard when getting close is my way, i can recommend it). And you must probably also revisit all parts of your surfing and remove any silly tweaks that helped boost your rating befor you where a precise surfer. Also try using a long BlindMansStick. It helps making it harder to wall segment against you and also makes you hit walls less often (the latter is a killer for precise prediction of course). I use 170. Makes for a less smooth wall movement, but I live with it. -- PEZ

I am using a wall smooth distance of 180 for the repcise reason you point out above. I have yet to evaluate the stop position. I think I will do that next. I need to think of an elegant way to do a three way comparison. Maybe I need to implement a similiar Move object and implement comparator. -- jim

I'm using precise prediction, a stick of 175, and I'm evaluating the stop position. Still stuck at 1970. You need to have pretty good prediction, but +- 10 pixels doesn't matter... I mean, how many people's "precise prediction" takes into account the fact that bots are implemented as non-rotating squares? I don't think that being off by a small amount matters much. --David Alves

You probably hava a bug or two somewhere. I don't take bot width into account at all. I let my bin smoothing take care of that. I think a prediction 10 pixels too far in any direction can make a world of difference when you're glued to the wall, smoothing hard and want to bail out. -- PEZ

That's true against head on targeters, where you can tell exactly where the enemy's bullet is and can therefore bail out perfectly if you can predict where you will be. But for targeters that aren't as straightforward, I don't think it's true... the less accurately you know where their bullets are, the less important it is to know exactly where you will be. Incidentally... isn't it like 5:30 in the morning there? Go to bed! :-) --David Alves


Robo Home | DarkHallow | Changes | Preferences | AllPages
Edit revision 8 of this page | View other revisions | View current revision
Edited October 12, 2004 23:31 EST by David Alves (diff)
Search: